P54 Presentation at 2007 ITU Fully Networked Car Workshop

24
Geneva, 7-9 March 2007 Using Speech to Interact with In-Car Devices in the Project54 System Andrew Kun University of New Hampshire

description

These are the slides from the presentation I gave at the 2007 ITU Fully Networked Car Workshop in Geneva, Switzerland.

Transcript of P54 Presentation at 2007 ITU Fully Networked Car Workshop

Page 1: P54 Presentation at 2007 ITU Fully Networked Car Workshop

Geneva, 7-9 March 2007

Using Speech to Interact with In-Car Devices in the Project54 System

Andrew KunUniversity of New Hampshire

Page 2: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

2Outline

o Introductiono Speech user interface testingo Driving simulator studieso Conclusion

Page 3: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

3What is the problem?

Page 4: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

4The system in the car

Page 5: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

5Speech user interface

o Command and control interfaceo Microsoft speech recognition (SR) engineo Microsoft text-to-speech (TTS) engineo Directional microphoneo Push-to-talk (PTT) buttono Grammars

Page 6: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

6Outline

o Introduction

o Speech user interface (SUI) testing

o Driving simulator studieso Conclusion

Page 7: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

7Testing

o Officer volunteers: 27o Corpus: just under 50,000 utteranceso Utterances: while PTT is pressed

Page 8: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

8Speech user interface (SUI) performance

o Recognized: 85 %o Unrecognized: 4 %o Misrecognized: 11 %

Page 9: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

9Reasons for imperfect recognition

o SR engine error: 37 %o User error: 63 %

Page 10: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

10SR engine error example

o Utterance: 0 3 2 1 8 5o → Recognized 0 3 2 1 0 5 7 1o → Recognized OK

Page 11: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

11User errors

o Utterance not in any grammar: 54 %o Utterance in another grammar: 34 %o PTT (“Patrol screen” ): 12 %

Page 12: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

12Outline

o Introductiono Speech user interface testing

o Driving simulator studieso Conclusion

Page 13: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

13Driving simulator

Page 14: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

14Driving simulator studies

o Multi-threaded dialogueso SUI and driving

Page 15: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

15Multi-threaded dialogues

o National Science Foundation granto Goal: interact with multiple real-time

devices using speecho Manual-visual task!o Human-human to human-computer

interaction

Page 16: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

16SUI and driving – police radio

Page 17: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

17SUI and driving

o Two experiments (8 subjects) :• Baseline + radio• Baseline + SUI

o Record:• Lane position• Velocity• Steering wheel angle

Page 18: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

18Lane position – radio

Page 19: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

19Lane position – SUI

Page 20: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

20Outline

o Introductiono Speech user interface testingo Driving simulator studies

o Conclusion

Page 21: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

21Status

o February 2007: ≈900 cars on the road in USA(New Hampshire, Massachusetts, California, Maryland)

o Industry participation

Page 22: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

22Research and development directions

o SUI and drivingo SUI performance improvements (SR

training, …)o SUI performance relation to driving task

difficulty, recognizer accuracyo Intelligent interaction (multi-threaded

dialogues, natural language processing, …)

o Non-speech work (handhelds, telematics, …)

o Standards

Page 23: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

23Acknowledgement

o US Department of Justiceo National Science Foundation

Page 24: P54 Presentation at 2007 ITU Fully Networked Car Workshop

The Fully Networked Car Geneva, 7-9 March 2007

24www.project54.unh.edu